Complete experience & project record

Data Scientist with an AI engineering track record.

I build forecasting, optimisation and agentic AI systems, backed by earlier experience across data engineering, full-stack products, deep learning research and decision-facing analytics.

This is the detailed record: where I worked, what I delivered, the technologies I used and the public evidence recruiters can follow.

10 professional rolesData science, engineering, product and research

20+ substantial buildsEnterprise systems, public projects and independent work

3 peer-reviewed papersApplied NLP, deep learning and intelligent systems

Current roleData Scientist · Complexica
Core positioningData Science × Applied AI
Based inAdelaide, Australia
Public workResearch · GitHub · Technical writing

Experience history

A career built across models, products and research

Each role includes the work itself—not just a title—and links to the organisation, publication, public project record or LinkedIn history where available.

March 2025 – present

Complexica · Adelaide, Australia

Data Scientist

Current

Building production ML and reliable applied AI systems for enterprise forecasting, optimisation and decision support. Current title held since January 2026.

  • Design human-in-the-loop AI support systems with evidence retrieval, explicit workflow state, deterministic verification, LLM evaluation and human approval.
  • Build agent harnesses around tools, context, memory, routing, retries, termination conditions and evaluator–optimiser loops.
  • Develop demand and promotional forecasting workflows using Prophet and gradient-boosting models with time-aware, leakage-resistant validation.
  • Apply causal inference, explainability and interaction modelling to understand why commercial metrics move and make results actionable.
  • Engineer automated Playwright quality gates for optimisation systems, including repeatability testing, negative controls and controlled fault injection.
  • Deliver full-stack ML and AI applications across AWS EC2, Amazon Bedrock and local environments using FastAPI, PostgreSQL, React, TypeScript and Docker.
  • Applied AI
  • Forecasting
  • Optimisation
  • Agentic systems
  • AWS
January 2024 – March 2025

SAcommunity / Infoxchange · Adelaide, Australia

Data Analytics Internship & Volunteer Information Officer

Completed a University of Adelaide industry internship and industry project, then returned as a volunteer in the Information and Data Analytics team.

  • Built Power BI, Google Analytics and Looker Studio reporting for South Australian community-service and council stakeholders.
  • Created geospatial dashboards for organisation discovery, visitor analysis and engagement trends.
  • Developed practical data-export and API workflows to work around Power BI and Google Analytics platform limitations.
  • Fine-tuned RoBERTa to extract venue-availability information from unstructured community descriptions into MARC-standard fields.
  • Co-authored and presented the resulting research at ALTA 2024, with publication in the ACL Anthology.
  • Python
  • Power BI
  • Looker Studio
  • RoBERTa
  • PyTorch
February – July 2024

ExpertEaseAI

AI Engineer & Consultant

Designed and built applied generative-AI prototypes for document, API and domain-specific question-answering workflows.

  • Architected multimodal retrieval-augmented assistants across text, document and API sources.
  • Built ingestion, preprocessing and cleaning pipelines to improve retrieval and response quality.
  • Integrated multiple model providers and open-source language models behind common application workflows.
  • Developed domain-specific evaluation and experimentation approaches without presenting prototype metrics as production results.
  • RAG
  • LangChain
  • Vector search
  • LLM APIs
  • AWS
May 2021 – April 2022

KH Analytics · Dhaka, Bangladesh

Data Engineer

Worked on enterprise data transformation and decision-support systems for a major retail engagement. Client identity and unsupported impact metrics are intentionally omitted.

  • Designed centralised analytical data structures spanning e-commerce, telesales and physical retail channels.
  • Developed Python, PySpark and SQL ETL workflows with automated data-quality checks.
  • Built Tableau and Power BI reporting for commercial and operational decision-making.
  • Contributed to customer segmentation, recommendation and pricing-oriented data products.
  • Python
  • SQL
  • PySpark
  • Snowflake
  • BI
November 2020 – April 2022

North South University · Bangladesh

Deep Learning Research Associate

Conducted applied deep-learning research across remote sensing, agricultural forecasting, economic estimation and intelligent systems.

  • Combined satellite imagery, environmental observations and machine learning for crop-yield and precipitation research.
  • Built transfer-learning workflows for estimating regional economic wellbeing from day and night imagery.
  • Developed experimental pipelines for feature extraction, statistical analysis and model evaluation.
  • Contributed to peer-reviewed IEEE conference publications and served as a verified journal reviewer.
  • Deep learning
  • Remote sensing
  • Time series
  • PyTorch
  • Research
February 2021 – March 2022

Antopolis · Bangladesh

Product Management & Strategy

Worked across product direction, platform requirements and delivery coordination for digital products.

  • Translated business problems into product requirements, workflows and delivery priorities.
  • Coordinated across product, engineering and stakeholder perspectives during implementation.
  • Applied analytics and technical understanding to product strategy and decision-making.
June 2020 – June 2021

Early Advantage · Bangladesh

Full Stack Software Developer

Built product and platform capabilities for a supply-chain finance network connecting smaller suppliers with corporate buyers.

  • Developed electronic-invoicing, inventory-management and dynamic-discounting workflows.
  • Built REST APIs and service integrations across operational platform components.
  • Created real-time analytics and notification capabilities for cash-flow and supply-chain visibility.
  • Contributed across backend, frontend, deployment and product delivery rather than a single application layer.
  • Django
  • Node.js
  • REST APIs
  • Analytics
  • CI/CD
March – October 2020

Beihang University

Research Engineer

Contributed to applied research engineering involving machine learning, experimentation and technical analysis.

  • Prepared and analysed research datasets for experimental machine-learning workflows.
  • Supported reproducible modelling, evaluation and technical documentation.
  • Worked across research questions that required both software implementation and analytical judgement.
February – August 2020

Space Research and Remote Sensing Organization · Bangladesh

Research Assistant

Applied geospatial analysis, remote-sensing data and machine learning to environmental and agricultural research problems.

  • Processed and analysed satellite and environmental datasets for research use.
  • Applied computer-vision and machine-learning methods to remote-sensing problems.
  • Developed features and experimental workflows connecting geospatial observations with downstream prediction tasks.
  • Documented findings and supported research communication across technical collaborators.
  • Remote sensing
  • Geospatial data
  • Computer vision
  • Machine learning
July 2019 – February 2020

Upskill · Bangladesh

Data Science & Machine Learning Mentor

Helped learners build practical foundations in data science and machine learning.

  • Delivered workshops spanning Python, data preparation, modelling and evaluation.
  • Guided learners through end-to-end projects rather than isolated algorithms.
  • Translated technical concepts into accessible explanations for mixed-experience audiences.

Project archive

A broader body of work

The archive includes current enterprise R&D, earlier professional builds, public research and independent projects. Employer-client names and internal implementation details remain anonymised.

Current applied AI & data science

Complexica · anonymised

Entity resolution · Retail data

Verified product matching pipelines

Two-stage matching systems for noisy retail product descriptions, combining deterministic constraints, TF-IDF similarity, local models and human review for ambiguous cases.

  • Brand and family matching before pack-size resolution
  • Hard filters to prevent plausible but invalid matches
  • Explicit “unmappable” outcomes instead of forced matches
PythonTF-IDFQwenDuckDB

Forecasting · Demand science

Promotional and seasonal forecasting

Leakage-safe forecasting and targeted demand corrections across promotional products, new-product instability and seasonal guardrails.

  • Time-aware evaluation on unseen periods
  • Prophet and gradient-boosting approaches
  • Model explainability and commercial interpretation
ProphetLightGBMXGBoostCatBoost

Applied AI · Optimisation

LLM-assisted promotion optimiser

A full-stack promotion-planning prototype where an LLM proposes plans and a deterministic layer recalculates every result before it can be accepted.

  • Cloud and local model backends
  • Classical solver retained for comparison
  • Deployed demonstration on AWS EC2
FastAPIReactCatBoostAWS EC2

Full-stack ML · Planner experience

Planner-facing forecasting platform

A forecasting application that lets planners explore a predicted outcome, understand its drivers and test changes against business targets.

  • On-demand feature and prediction services
  • Explainable planning interface
  • Containerised AWS delivery
FastAPIPostgreSQLReactDocker

Pricing intelligence · Data quality

Competitor pricing intelligence

A pipeline that turns large, messy retail-price scrapes into defensible competitor pricing through constrained matching, live checks and structural data-quality audits.

  • Large-scale processing with DuckDB
  • AI tie-breaking behind deterministic constraints
  • Evidence-led challenge of business assumptions
PythonDuckDBLLMsStreamlit

Agent infrastructure · Local LLMs

Local AI research environment

An always-on environment for agent development, local inference and automated testing, supported by reusable runbooks, process supervision and safe model-routing patterns.

  • Qwen served through LM Studio
  • Reusable agents, hooks, MCP and memory strategies
  • Local-first routing for cost and privacy
LM StudioMCPClaude CodePostgreSQL

Detection · Decision intelligence

Early-warning and churn detection

Production-minded detection systems that transform behavioural and operational data into interpretable leading indicators for intervention.

  • Feature engineering and validation pipelines
  • Risk scoring and driver interpretation
  • Decision-facing outputs for business teams
PythonGradient boostingExplainabilityFastAPI

Optimisation · Technical audit

Solver and price-curve audits

Architecture and algorithm audits that traced structural constraints, memory pressure, stale calculations and modelling assumptions across optimisation components.

  • Reverse-engineered multi-process solver flows
  • Identified algorithmic and data-contract weaknesses
  • Translated findings into prioritised engineering recommendations
JavaOptimisationJMSSystem analysis

Architecture · ML delivery

Agent and model-serving architecture

Design work for tool-using agents and production ML services, covering state machines, validation engines, model runtimes and deployment trade-offs.

  • Plan–Act–Observe–Reflect workflows
  • FastAPI, sidecar and embedded-runtime comparisons
  • Discovery-led architecture decision matrices
FastAPIDockerONNXReact

Earlier, public & independent work

Research · products · community

Applied NLP · Published research

MARC-standard information extraction

Fine-tuned RoBERTa to transform unstructured descriptions of community venues into structured MARC fields, supported by preprocessing, annotation and evaluation workflows.

ACL Anthology publication ↗

Analytics · Community impact

South Australian council analytics

Data pipelines and interactive reporting for engagement trends, service discovery and geospatial exploration using Google Analytics, Power BI and Looker Studio.

Public project record ↗

Enterprise AI · Independent

ExpertEaseAI multimodal assistant

A retrieval-augmented assistant combining document and API ingestion, preprocessing, vector search and multiple model providers for domain-specific question answering.

LangChainChromaHugging FaceAWS

Data engineering · Retail

Enterprise data transformation

Centralised data architecture, ETL workflows, data-quality controls, business intelligence and ML-enabled retail products for a large retail engagement.

SnowflakePySparkSQLPower BI

Deep learning · Remote sensing

Crop yield and precipitation forecasting

Deep Gaussian process and time-series research combining remote-sensing data with deep learning for agricultural forecasting in Bangladesh.

IEEE publication ↗

Computer vision · Economic analysis

Satellite-based wellbeing analysis

Transfer-learning workflows using day and night satellite imagery to investigate whether visual and geospatial features can support regional economic estimation.

VGG19PythonGeospatial analysisTransfer learning

Fintech · Full-stack product

Early Advantage supply-chain finance platform

Electronic invoicing, inventory management, dynamic discounting and analytics designed to improve working-capital visibility across supplier and buyer workflows.

DjangoNode.jsREST APIsChart.js

Agentic AI · Public code

AI fitness agent with LangGraph

A documented agent implementation demonstrating state management, tool use, error handling and Amazon Bedrock integration through a practical fitness-assistant workflow.

Computer vision · Full-stack

Facial-recognition attendance system

A real-time attendance and member-management application combining facial recognition, entry and exit logging, digital invoicing and operational analytics.

Django RESTMTCNNVue.jsOpenCV

LLMs · Document intelligence

Conversational PDF question answering

An earlier document assistant that let users upload PDFs, request summaries, find references and ask questions using retrieval and language-model components.

TransformersspaCyOpenAIPyTorch

IoT · Intelligent systems

EZ Parking smart reservation system

An intelligent parking-space reservation concept combining connected devices, availability management and a user-facing booking workflow.

IEEE publication ↗

Machine learning · Growth analytics

Customer acquisition optimisation

Customer-segmentation, campaign experimentation and lifetime-value modelling used to compare acquisition strategies and improve marketing decision support.

PythonClusteringA/B testingPredictive modelling

Data systems · Decision support

Real-time query and decision-support systems

Earlier work on multi-source data integration, query processing and automated response workflows for operational decision support.

PythonSQLStreamingData integration

Peer-reviewed research

Published work recruiters can verify

Three conference papers show a consistent applied research thread across natural language processing, remote sensing, deep learning and intelligent systems.

2020

IEEE International Geoscience and Remote Sensing Symposium

A Deep Gaussian Process for Forecasting Crop Yield and Time Series Analysis of Precipitation Based in Munshiganj, Bangladesh

Applied deep learning, remote sensing and agricultural forecasting
View publication ↗
2018

International Conference on Advanced Computer Science and Information Systems

EZ Parking: Smart Parking Space Reservation Using Internet of Things

Applied IoT and intelligent mobility systems
View publication ↗

Writing & knowledge sharing

Making technical work understandable

Writing is part of the engineering work: it forces assumptions into the open and makes systems easier to evaluate, reuse and improve.

View all writing on Medium ↗

Connect

Current depth, with the history to support it.

For the concise version of my current positioning, return to the main portfolio. For a conversation about applied AI, data science or a specific project, connect with me directly.